chemical detection using the airborne thermal infrared imaging

The detection of most organic, organophosphates, and similar high molecular weight compounds is best accomplished in theIR (see typical spectral assignments in Fig. 1). There are several challenges that must be addressed in order to accomplishthese airborne IR remote sensing measurements: (i) the intervening atmosphere significantly affects the spectral signaturesobserved by the airborne sensor, (ii) the plume-to-ground temperature contrast is often very small and may be positive ornegative, (iii) the background terrain that acts as an IR source has unknown spectral properties, and (iv) real-timeconsiderations require a short exposure time and fast signal processing.

Figure 1. Typical spectral assignments of some organic compounds and the atmospheric “windows.”

2. APPROACH

2.1. The Sensor

The approach selected to address these problems is based on a low spectral resolution sensor (relative to FTIR spectrometers)specifically developed for this application. The Thermal InfraRed Imaging Spectrometer (TIRIS) is an airborne instrumentunder development for emergency response, the detection of toxic compounds, air pollutants, chemical warfare agents, and fortracking their plumes in real-time. The TIRIS operates in 20 spatial and 64 spectral bands between 7.5 to 14.0 µm at ∆λ ≈ 0.1 µm

resolution (∆ν ≈ 10 cm-1 at λ=10 µm), in a down looking mode, and detects / identifies plume composition against the thermalradiation of the background terrain. A laboratory prototype, TIRIS-I was constructed and tested. A field deployable version,TIRIS-II in now under development [1]. Being an imaging spectrometer, the plume is simultaneously characterized spatially andspectrally, and the measurements can be registered to a map location via geographical information systems (GIS).

The benefit of a low spectral resolution, dispersive system, is that it allows the capture of a relatively broad spectral signature(7.5 to 14.0 µm in this case) in a single short-exposure “snapshot.” In order to ensure that the resolution of TIRIS is stillsufficient to detect target gases, the entire EPA gas phase library of the most hazardous air pollutants (HAPs) was deresolved

SPIE, Orlando, FL, April 1997

Pg. 2

from its original 0.25 cm-1 resolution to an equivalent 100 nm resolution of the TIRIS. A comparison of one set of original andderesolved spectra, shown in Fig. 2, indicates that although the fine structure of the gas phase spectrum is lost, the overallstructure is clearly adequate to detect and in most cases identify the gas.

Figure 2. TIRIS resolution is sufficient to characterize IR spectrum. The top line corresponds to transmission measured by anFTIR instrument in a vapor phase cell at 0.1 cm-1 resolution. The lower line is the absorption cross section calculated from the

transmission data and deresolved to 100 nm, or about 10 cm-1 at 10 µm.

2.2. Sensor Model: Forward Solution

A phenomenology model (forward solution) that serves as a sensor design tool has been developed to simulate the sensorresponse for a better understanding of its operational parameters and limitations. This is a radiative transfer model thataccounts for the terrain infrared characteristics, sky radiation, and the plume parameters (Fig. 3). A plume may be observed inabsorption or emission depending on the temperature contrast with the background. Several spectral libraries, of organiccompounds and of background materials are used to model and predict the sensor response under a prescribed scenario. Thelibraries used for the modeling include the EPA's HAPs spectral library, the EPA/NIST vapor phase IR library, and the NPICspectral library of background materials. Other libraries can be incorporated into the model, and the model itself can beadapted to other sensors. A list containing Internet sites of relevant spectral libraries can be found on the WWW[2].

The model, depicted in Fig. 3, shows the various sources of radiation and the exchange of radiation in the measurementenvironment. The model takes into consideration the terrain spectral characteristics, and any exchange of radiation with otherelevated terrain features or sky and clouds. The ground can be assumed to behave as a Lambertian black or gray bodysource, or a real source based on selected elements from the backgrounds library. The plume is modeled under the thin gasapproximation as a mixture of any number of compounds. Their mixing proportions can be specified.

In general, the radiation measured by the sensor along a line of sight to the ground can be described by Eqs. 1 and 2, wherethe nomenclature corresponds to that in Fig. 3. Eq. 1 expresses the radiation directly emitted by the terrain, plus reflected sky


Pg. 3

radiation. Eq. 2 expresses the direct emission from the plume and absorption of terrain radiation as it passes through theplume.

Nλ,b = αλ $ 2hc2 / {λ5 [exp(hc/λkTb ) - 1] } + (1 - αλ )$ 2hc2 / {λ5 [exp(hc/λkTS ) - 1] } (1)

Nλ,r = Nλ,b $ exp (-kλL) + Bλ(λ,Ta) $ [1-exp(-kλL)] (2)

In these equations, Tb, Ts, and Ta are the temperatures of the background terrain, the sky (e.g., clouds), and the plume,respectively. k is an absorption coefficient (a function of wavelength, λ ), and L is the path length through the plume. Theemissivity α is also a function of wavelength.

A computer code that incorporates the model with a graphical user interface has been developed and a screen capture isshown in Fig. 4. In this forward solution the user can “create” synthetic mixtures of various library compounds, specifyplume, atmospheric and ground parameters. The model calculates and displays various spectra of interest such as thecombined absorption cross section signature of the selected mixture (bottom left of Fig. 4), the radiance at the sensor aperture(bottom right), or the sensor output (using the sensor transfer function). All calculations are conducted at 1 cm-1 intervals.Sensor output is given at the specific sensor resolution (100 nm for TIRIS).

Figure 3. IR radiative exchange model for airborne plume detection.

The spectral signature at the aperture of the sensor is used here to illustrate the effect of the thermal contrast. In thisparticular case the terrain is warmer than the plume by about 10 °C. The Planck function at the terrain (Tb) and plume (Ta)temperatures are depicted in the figure (bottom right). The radiance at the sensor can be seen confined between these two


Pg. 4

curves. As the temperature contrast becomes smaller, the plume signature diminishes to zero at the reversal point. Once theplume becomes hotter than the terrain it reappears this time in emission (still confined between the two curves correspondingto the Planck blackbody at the gas and terrain temperatures).

As depicted in Figs. 3 and 4, and expressed by the term Tb (background temperature) in Eq. 1, terrain background emission isan important factor in the observed signature. Since this is seldom known it has to be estimated from the actual sensor data.To validate an approach for extracting ground emissivity and temperature (addressing the so called ε-T problem), a specialexperiment was conducted. The ATLAS multispectral scanner, operated by NASA Stennis Space Center, was used to collectIR signatures over Baton Rouge, LA. The spectral signatures from the six thermal IR bands of ATLAS were used then in aniterative procedure to extract ε-T information. A false-color map of the terrain apparent temperature is shown in Fig. 5[3]. TheATLAS technique was found sufficiently accurate and has been ported to the TIRIS sensor environment.

Figure 4. Radiative exchange and sensor response model. The plot on the bottom right shows that the expected sensorsignature is located between two curves that depict the Planck blackbody radiance at the ground and the plume temperatures.

The forward solution was written in IDL (interactive data language) that runs on all the popular computing platforms, and canbe incorporated into ENVI, a hyperspectral data processing package. Atmospheric correction are obtained via theMODTRAN code that was compiled to run on a PC using the 32-bit version of Microsoft Fortran 90 and MS developmentstudio.

2.3. Sensitivity Estimates

The radiative transfer model discussed above applies to monochromatic radiation only. In attempting to estimate detectionsensitivities or reduced broadband absorption spectra to determine species column densities, one needs to know whether


Pg. 5

spectrally averaged values of the absorption coefficient kλ can be used. Such an assumption is only valid if the averagespacing between the spectral lines in the spectral band being used is less that the typical line width[4]. Spectral line widths areusually about 0.1 - 0.2 cm-1 at 1 atmosphere. When these conditions apply (as they usually do for molecules with more thanabout five atoms at one atmosphere pressure), one can apply the so-called "smeared line model" and use average absorptioncoefficients. Absorption based upon use of average spectral absorption coefficients will be less than predicted by Eq. 2 if thespectral lines do not overlap.

The smeared line model can also be used in weak concentration limit where kλ L<< 1 at all wavelengths, including the centersof strong spectral lines. In this case, the spectral lines are all "optically thin" and as a result exponential terms in the radiativetransfer equation can be linearized.

Preliminary detection sensitivity estimates in the following paragraphs are based on the smeared-line model whererepresentative values of the average spectral absorption coefficient per molecule in the LWIR bands are needed to estimatespecies detection thresholds. OKSI used the EPA's HAPs spectral library to obtain average spectral absorption coefficients.The EPA library, obtained using an FTIR spectrometer, is measured in 0.25 cm-1 intervals; average values were calculated overrelatively broad spectral intervals (100 nm -- corresponding to the TIRIS sensor), and dividing by the column density in themeasuring device. The resulting absorption cross-section values range from 200 to 10,000 m2/kmole for strong bands,corresponding to an absorption cross-section of 3x10-21 to 2x10-19 cm2/molecule. These results are in reasonably goodagreement with the theoretical band-integrated absorption coefficients (10-19 to 10-17 cm-1/moleclue) divided by the typicalvibration-rotation band width of 10 to 100 cm-1 (IR band width tend to fall in a narrow range governed by the temperature andthe molecular moment of inertia) which result in a range of LWIR peak absorption cross sections from 10-21 to 10-18

cm2/molecule.

Therefore, we assume

10-21 < kλ /ni < 10-18 cm2 /molecule (3)

for representative LWIR bands, near band center. Note that if the bandpass ∆λ ≈ 100 nm or ∆ν ≈100 cm-1 is used for thedetection instrument, as with the TIRIS, the average absorption coefficient over the band might be considerably less, since theabsorption band would be narrower than the measurement bandpass.

Next we set a detection "threshold" of say A ≈ 5% , for high-confidence detection of particular species. Here we define A tobe the relative reduction in radiance along the ray path from the surface to the receiver. We will also assume that Tb - Ta =10°C, so that according to Eq. 2,

A = [ Nλ,b - Nλ,r ]/Nλ,b ≈ 0.15 $ [ 1 - exp(-kλL) ] (4)

Eq. 2, with minor modifications, can also be used to estimate the value of the factor that precedes the [ 1 - exp(-kλL) ] term forother values of Tb - Ta . If one sets a minimum absorption threshold at A=0.05, the corresponding required value for the [ 1 -exp(-kλL) ] term is A/0.15 = 0.33, which requires

kλL > 0.4 (5)

for high probability of detection. Now assuming 10-21 < kλ /ni < 10-18 cm2 /molecule , this implies that the column density niLmust satisfy the inequality

niL > 4x1017 to 4x1020 molecules/cm2 (6)

If the absorbing species "i" that one hopes to detect is confined to a convective mixing layer near the ground that is 10 metersthick, the corresponding species detection limit would be

ni > 4x1014 to 4x1017 molecule/cm3 (7)


Pg. 6

Eq. 7 corresponds to species mole fraction in a 0.001% to 1% range, approximately. More accurate estimates of requiredcolumn densities for detection can be made once the species are defined, temperature difference Tb-Ta are known, and aminimum detection threshold level of A required for high-confidence species identification is established. The forwardsolution computer model is used to generate such estimates.

2.4. Plume Detection: Inverse Problem Solution

The second model (inverse solution) uses the sensor output to determines the composition of organic gases that created theobserved IR signature. The inverse solution has two steps. First, the organic compounds library is scanned and reduced to aset of M potential end-members that may contribute to the signature, based on detection probability and constant false alarmrate considerations. In the second step, an MxM matrix is constructed in which the M compounds selected in the first step areincorporated into M linear equations, in M spectral bands. In practice, M<< # of bands. The solution of this system ofequations provides an exact solution to the problem rather than the traditional least squares approximation. The key to thisapproach is the proper selection of the M band.


Pg. 7

Figure 5. False color terrain temperature map obtained From ATLAS IR imagery.

The validation of the inverse model suffers from the lack of real data with accurate ground truth. While developing the code,synthetic data were produced from the spectral libraries (using the forward solution). These data were then “corrupted” withvarious models of noise and various levels of SNR. Results of one trial, summarized in Fig. 6, show 100% detection success.Work underway presently with field data will be used to validate these algorithmic paradigms.


Pg. 8

Figure 6. Example of Inverse Solution Analysis Results. In this case, synthetic mixtures were created from anarbitrary number of compound in the list. The Inverse solution was able to determine which compounds

actually were included and at what composition.

0.0000E+000.0E+001,2-Epoxybutane

0.0000E+000.0E+00Dichloroethyl ether

4.9877E-045.0E-041,2-Dibromo-3-chloropro

0.0000E+000.0E+00m-Cresol

0.0000E+000.0E+00Chloroprene

0.0000E+000.0E+002-Chloroacetophenone

0.0000E+000.0E+001,3-Butadiene

1.0003E-031.0E-03Benzyl chloride

Inverse Solution EstimatedTrue ConcentrationEnd-Member

PERFORMANCE OF INVERSE SOLUTION

3. LABORATORY VALIDATION

The TIRIS-II sensor is currently set up in the laboratory for calibration and characterization, as depicted in Fig. 7. A vacuumcell with KBr windows was designed to provide the required optical path length. Various vapor phase compounds may beinjected into the cell in controlled amounts. The cell temperature may be raised above ambient and the pressure accuratelymonitored to a subtorr level, from which the column density of the vapor can be calculated. The source of IR backgroundradiation is a blackbody that can vary over a wide temperature range. The combination of background and cell temperaturepermits observation of gas spectral signatures either in emission or absorption. Background spectra obtained using only theevacuated cell and blackbody can be used to obtain the overall system transfer function (including FPA, various filters, andwindows spectral characteristics). The laboratory tests are currently underway to validate both the forward and inversesolutions.

Gas Cell

Off-Axis Parabola

Off Axis parabola

TIRIS Sensor

with KBrWindows

Cold Stop

FPA

ZnSe Window

BlackbodySource

Figure 7. Schematic of Laboratory Setup for the TIRIS Vapor Phase Detection Characterization and Sensor Calibration.

4. SUMMARY

Extensive use of synthetic data was made for the validation of the algorithms. In order to futher validate the approach to dataanalysis, OKSI is studying spectral signatures collected with an FTIR airborne Atmospheric Emission Spectrometer. Thesespectra, measured at a higher spectral resolution than the TIRIS, were obtained from an altitude of 5,000 meters, and cover a


Pg. 9

plume emitted by a coal burning power plant and of the terrain in the vicinity of the plant. Data processing is currentlyunderway, aided by extensive ground truth that are available for this data set.

ACKNOWLEDGMENT

The work described here was performed under the following contracts: DARPA # DASG60-94-C-0088, U.S.A.F #F08637-95-C6017, and NASA # NAS5-32935.

REFERENCES

[4]S.S. Penner, "Quantitative Molecular Spectroscopy and Gas Emissivities." Addison Wesley, Reading MA 1959 (Chapter 8).

[3]N. Gat and W.D. Graham, “Airborne Digital Thermal Imagery for Remote Sensing of the Environment.” Earth ObservingMagazine, Aug. 1995, pp. 51-53.

[2]See Directory of Imaging Spectroscopy Resources on the WWW at http://www.techexpo.com/opto-knowledge

[1]N. Gat, et-al, “Thermal Infrared Imaging Spectrometer (TIRIS) Status Report.” Infrared Technology & Applications XXIIISPIE Vol. 3061, 1997


Pg. 10

Chemical Detection Using the Airborne Thermal InfraredImaging Spectrometer (TIRIS)

Nahum Gata, Suresh Subramaniana, Jacob Barhenb, Mike Sheffielda and Hector Erivesa

a Opto-Knowledge Systems, Inc. (OKSI), 1737 Third St., Manhattan Beach, CA 90266 b Oak Ridge National Laboratory, Oak Ridge, TN 37831

ABSTRACT

A methodology is described for an airborne, downlooking, longwave infrared imaging spectrometer based technique for thedetection and tracking of plumes of toxic gases. Plumes can be observed in emission or absorption, depending on the thermalcontrast between the vapor and the background terrain. While the sensor is currently undergoing laboratory calibration andcharacterization, a radiative exchange phenomenology model has been developed to predict sensor response and to facilitatethe sensor design. An inverse problem model has also been developed to obtain plume parameters based on sensormeasurements. These models, the sensors, and ongoing activities are described.

Keywords : organic compounds, chemical warfare agents, drug manufacturing byproducts, imaging spectrometer, plumetracking, infrared, terrain thermal emission, inverse problem, TIRIS.

1. THE PROBLEM

When toxic compounds are spilled, either following an accident in a chemical plant, derailment, trucking accident, etc., toxicplumes may drift or be wind-blown into residential areas. The emergency response of authorities is to evacuate large areas. Insome recent cases where uncontrolled fires have burned in chemical plants, residents had to stay out of their homes forseveral days, as a safety measure rather than due to actual presence of toxic fumes. Such situations point out the need forremote sensing techniques for evaluation of the hazard levels. Similarly, clean up crews can use remote sensing techniques toassess the hazard level before attempting to access the affected area. Presently, they must approach the contained area inprotective gear and use gas sampling based sensors (or the canary in a cage method) to detect potential hazardous material.Single point measurements, however, are of lesser value than line-of-sight data for planning strategies of egress or cleanup.

The present state-of-the-art in IR spectral remote sensing is based on open air-path Fourier transform infrared (FTIR)spectrometers. Typically, a monostatic or bistatic arrangement is used where a hot source is located across the field ofmeasurement (bistatic) or adjacent to the sensor with a retrorefletor across the field (monostatic). The source, at temperaturemuch higher than that of the gas plume, emits infrared (IR) radiation allowing the plume to be observed in absorption. Whenplumes are released from a stack at an elevated temperature, they can be observed by FTIR instruments in emission and thereis no need for a IR source. Emission measurements are often performed with the (cold) sky as a background. The sky radianceis typically low relative to that of the hot plume and the signal to noise ratio (SNR) is acceptable for reliable measurements.

In an emergency response situations, however, the placement of retroreflectors or IR sources, or having a specific view anglefor the plume is not always practical, considering that the wind directions keep shifting. Moreover, the extent of plume spreadis typically unknown and a wide area search may be required to continuously delineate the boundaries of the plume spread. Insuch situations airborne sensors that operate in the downlooking mode may be useful. The purpose of this paper is todescribe the status of an ongoing effort to address the technical challenge of such a detection scheme.


Pg. 1

Further author information: OKSI: Phone: 310/371-4445; E-mail: [email protected]; URL: http://www.techexpo.com/opto-knowledge J.B.: Phone: 423/574-7131; E-mail: [email protected]

The detection of most organic, organophosphates, and similar high molecular weight compounds is best accomplished in theIR (see typical spectral assignments in Fig. 1). There are several challenges that must be addressed in order to accomplishthese airborne IR remote sensing measurements: (i) the intervening atmosphere significantly affects the spectral signaturesobserved by the airborne sensor, (ii) the plume-to-ground temperature contrast is often very small and may be positive ornegative, (iii) the background terrain that acts as an IR source has unknown spectral properties, and (iv) real-timeconsiderations require a short exposure time and fast signal processing.

Figure 1. Typical spectral assignments of some organic compounds and the atmospheric “windows.”

2. APPROACH

2.1. The Sensor

The approach selected to address these problems is based on a low spectral resolution sensor (relative to FTIR spectrometers)specifically developed for this application. The Thermal InfraRed Imaging Spectrometer (TIRIS) is an airborne instrumentunder development for emergency response, the detection of toxic compounds, air pollutants, chemical warfare agents, and fortracking their plumes in real-time. The TIRIS operates in 20 spatial and 64 spectral bands between 7.5 to 14.0 µm at ∆λ ≈ 0.1 µm

resolution (∆ν ≈ 10 cm-1 at λ=10 µm), in a down looking mode, and detects / identifies plume composition against the thermalradiation of the background terrain. A laboratory prototype, TIRIS-I was constructed and tested. A field deployable version,TIRIS-II in now under development [1]. Being an imaging spectrometer, the plume is simultaneously characterized spatially andspectrally, and the measurements can be registered to a map location via geographical information systems (GIS).

The benefit of a low spectral resolution, dispersive system, is that it allows the capture of a relatively broad spectral signature(7.5 to 14.0 µm in this case) in a single short-exposure “snapshot.” In order to ensure that the resolution of TIRIS is stillsufficient to detect target gases, the entire EPA gas phase library of the most hazardous air pollutants (HAPs) was deresolved


Pg. 2